Codex reset, usage limits, and banked resets FAQ | WHENRESET

In plain terms, it means you get more usable Codex capacity. Your current usage allowance may be restored immediately, or OpenAI may give you a banked reset that you can save and activate later. Both let you keep using Codex, but they are not the same event.

Example:Suppose your weekly allowance is at 9% and would normally recover on October 2 at 10:00. If OpenAI restores the allowance today, it returns to 100% now. If OpenAI gives you a banked reset instead, it stays in your account until you choose to use it.

Note:WHENRESET groups both under reset events, while each event is still labeled as an allowance reset or a banked-reset grant.

First confirm whether the announcement refers to a direct reset or the issuance of a reset card, then confirm whether your plan, account, and workspace meet the conditions. If what was issued is a banked reset, the card arriving does not mean the limit has already been refreshed; if it is an automatic reset, it also only applies to the accounts and limits covered by the announcement. Please refresh "Settings → Usage" and confirm that you are logged in to the correct account and workspace. If you meet the conditions but it still has not arrived after a long time, contact OpenAI Support for investigation.

Example:The announcement says, "Eligible Pro users will receive 1 banked reset." Plus users will not receive a card because of this; after eligible Pro users receive the card, if they do not use it manually, the current weekly limit will not automatically return to full.

A direct reset immediately refreshes the limits covered by the announcement; a reset card is first stored in the account and must be used manually by you. What people often call a "reset card" is officially called a banked reset. Fully using one banked reset refreshes Codex's 5-hour and weekly windows and changes the weekly recovery date; simply receiving the card does not automatically change your current limits.

Example:Suppose your weekly limit is already used up. After a direct reset is completed, the usage page will show that the limit has been refreshed; if what you received is 1 banked reset, your current limit may still be 0, and only the "available resets" count changes from 0 to 1. You go to the usage page, click to use it, and confirm, and only then is the limit actually refreshed.

One is saved for later, one restores capacity immediately after payment, and one charges for usage after the included allowance is exhausted. A banked reset usually comes from a promotion or compensation and stays in your account until you activate it. A paid instant reset restores the relevant allowance immediately after purchase. Extra usage lets you continue after the included allowance is exhausted and charges for what you use. Your account may not offer all three.

Example:Think of a banked reset as a spare battery, a paid reset as buying a full battery now, and extra usage as connecting a power bank that charges by consumption.

Note:Availability, pricing, and expiry depend on what your own account currently shows.

You cannot assume they will be reissued; eligibility depends only on the deadline and conditions stated in that campaign. There is no long-term official rule that "after upgrading, all historical reset cards are automatically made up." Each campaign can have different requirements for plan, region, account status, creation time, or subscription deadline. After upgrading, look only at the current campaign description and the available count shown on your usage page.

Example:Suppose the July 1 campaign required "already subscribed to Pro before the campaign started that day," but you upgrade to Pro only on September 1. You cannot demand to receive this card just because it was issued historically; if a new September campaign explicitly includes newly upgraded users, then follow the new campaign.

Consider it when your weekly allowance is nearly exhausted, natural recovery is still far away, and you must keep working now. If plenty of allowance remains or it will recover tomorrow, using a reset is usually poor value. The whole reset is consumed; the unused portion is not saved for another activation, and the old recovery time may be recalculated.

Example:Suppose 85% remains and the allowance will naturally return to 100% tomorrow at 10:00. If you wait, you can use today’s 85%, recover for free tomorrow, and keep the reset. If you activate it now, you spend the whole reset just to add 15%. Even if you use only another 5% today, you should not assume the old recovery will still happen tomorrow. That can waste the reset.

Note:Before activating, open Settings → Usage and compare the remaining percentage with the natural recovery time.

That promotion required the reset to be used within 30 days. This does not mean every future banked reset has a 30-day life. For the June 2026 referral promotion, a banked reset stayed in the account but had to be activated within 30 days. Expiry did not automatically activate it. Other promotions, compensation grants, or future resets may use different rules.

Example:If you received that promotional reset on June 15, you could not treat it as permanent stored credit. You needed to decide before the expiry date shown in your account.

Note:Always follow the expiry shown in your account and do not generalize one promotion’s rule to every reset.

Reset cards are not issued on a fixed monthly basis, and not all Plus and Pro users will automatically receive them. Saveable resets usually come from the current promotion, invitation rewards, or official temporary activities. Each activity's plan, region, account status, invitation conditions, and deadline may differ; whether you are eligible depends only on the current activity description and your own "Settings → Usage." Third parties cannot claim it for you, nor can they guarantee supplementary issuance.

Example:A friend's Pro account received 1 reset because they completed the current invitation task, but that does not mean your Pro account will automatically show one too. If your page has no "available reset," first check the current activity conditions rather than assuming the system missed sending it.

First verify eligibility, the correct account, and workspace, then check whether the card was already used or has passed the expiration time shown on the page. Distribution may only cover eligible accounts; reset cards may also have their own expiration time. Please refresh or reopen the usage page, confirm that the logged-in account, workspace, and plan are correct, and check the available count and expiration time. If the page previously clearly showed a card that had not been used but disappeared early, save a screenshot, the time, and the time zone, and contact OpenAI support for investigation.

Example:You see 1 reset on your personal Pro account, but later when you switch to a Business workspace it shows 0. This does not necessarily mean the card is lost; first switch back to the original account and workspace to confirm.

It was a limited-time promotion that has ended, not a permanent referral feature. The promotion ran from June 11 through June 24, 2026. Referrers and invitees had to satisfy the account, plan, and completion requirements shown on that promotion page. Those terms should not be treated as a referral program that is still active now.

Example:Finding a June referral link today does not mean that registering or upgrading through it will still award a reset, because the promotion period has ended.

Note:If a future referral promotion appears, follow the dates and requirements on the new promotion page.

Watch both. Whichever is exhausted first can limit you first. The five-hour limit controls usage over a short window, while the weekly limit controls total usage across the longer cycle. Remaining capacity in one does not guarantee access if the other has been exhausted.

Example:Your five-hour allowance can still show 80% while the weekly allowance is exhausted, so Codex may still be unavailable. The reverse can also happen after intense short-term use.

Note:Open Settings → Usage and check both meters and their separate recovery times.

Do not keep relying on the old recovery time. Check Settings → Usage again after using the reset. A banked reset restores the relevant allowance and may recalculate the next recovery time, so the date shown before activation may no longer apply.

Example:If Usage originally said September 26 at 10:00 and you activate a reset on September 25 at 14:00, do not assume the allowance will reset again the next morning. Use the new date displayed in Usage.

Note:WHENRESET cannot read your account, so it cannot replace the exact recovery time shown there.

The same message does not necessarily consume the same amount of quota, and Pro may also have different usage tiers. First confirm whether both of you are on the same Pro tier, account, and workspace. Then compare the model, reasoning effort, speed mode, context length, where the task runs, tool calls, and duration. The official information only provides approximate usage ranges and does not promise a fixed "amount deducted per message."

Example:You both send 10 messages: your friend only edits a few lines of text, using a lighter model and standard speed; your 10 tasks need to read the entire repository, run tests, repeatedly fix issues, and use high reasoning effort. The message count is the same, but your actual consumption may still be much higher.

Not all of them. ChatGPT Work and Codex share usage; API billing is separate; regular ChatGPT conversations have their own model and message limits. Work performed in ChatGPT Work and Codex should be understood through OpenAI’s shared-usage rules. API calls use separate API billing and limits. Regular ChatGPT conversations also have their own model and message limits, so these should not be treated as one meter.

Example:Heavy Codex use can reduce the shared usage you see when you continue in ChatGPT Work. Calls made with an API key, however, are billed through the API account rather than deducted from that shared meter as API credit.

Note:Limits can change by plan and product. Check your current Usage page and API billing page.

No. The post time is only the time the news was made public, not the time it reaches your account, and it is not necessarily the start of a new cycle. First check whether the post is talking about an automatic reset, the issuance of a saved reset, or some other event. An automatic reset refreshes the quota covered by the announcement; a saved reset waits until you manually use it; for a paid instant reset, the new weekly cycle starts from the first request after the reset. For other types, whether the recovery date changes should be based on the usage page after the reset.

Example:Tibo posts at 09:00 Beijing time, and the system applies the reset to your account only at 09:40. You cannot directly treat 09:00 as the start of your personal cycle; if the usage page shows a new recovery time after 09:40, use the page time.

Open "Settings → Usage"; when using Codex CLI, you can also enter /status. The Usage page shows the currently applicable 5-hour and weekly windows, reset times, extra usage balance (Credits), and whether a reset is available. When reading percentages, first confirm whether the label means "used" or "remaining." Third-party sites can only provide public events and predictions; they cannot read your personal account quota.

Example:WHENRESET shows "a public reset may occur tonight," but your Usage page clearly says "weekly quota resets September 26 at 15:00." Until a new official event actually applies to your account, when your personal quota resets should be based on this page.

There is no unified “Monday at midnight every week”; the exact time is based on what your own usage page or /status shows. The weekly quota is an account-specific time window, not a calendar week shared by all users. After using a full reset, purchasing an instant reset, or participating in certain official reset events, the recovery date may also change. Therefore, other people's weekday, time zone, and countdown cannot be directly applied to your account.

Example:A friend shows recovery at Monday 10:00 Beijing time, while you show recovery at Thursday 18:20. Both can be correct at the same time. You should arrange your usage according to Thursday 18:20 on your own page.

There is no single fixed message count or token number that is accurate for every user. Consumption depends on the model, task length, context, tool calls, plan, and other factors. Two requests that both look like one message can consume very different amounts, so a universal message count cannot reliably predict remaining usage.

Example:A request to edit one line and a request to inspect an entire repository, run tests, and change several files may each appear as one message, but they do not consume the same amount.

Note:Use the live meter in Settings → Usage to judge what remains.

A full reset or paid instant reset may reschedule the weekly window; also first rule out looking at the wrong account or workspace. First recall whether you used a saved reset, purchased an instant reset, or encountered an official automatic reset, then check the current account and workspace. Fully using a saved reset refreshes the 5-hour and weekly windows and changes the weekly recovery date; the new weekly window after a paid instant reset also starts over. In the end, only the latest time shown on the usage page is authoritative.

Example:The page originally said recovery on Saturday. You used a full reset on Wednesday, and afterward it changed to recovery around the following Wednesday. This is usually not a broken countdown, but rather the weekly window having been rescheduled.

Check whether your account offers a banked reset, a paid instant reset, or extra usage. If it offers none of them, wait for natural recovery. Available options differ by account. Open Settings → Usage to see the recovery time and whether a banked reset, paid reset, or extra-usage option is present. Do not assume an option is available if your account does not show it.

Example:If recovery is two hours away and no reset or extra-usage option appears, waiting is the safest choice. If you do have one banked reset, decide whether saving two hours is worth consuming the entire reset.

Note:WHENRESET never purchases or activates a paid feature on your behalf.

Not necessarily. A task that has already started may continue to completion within fair use limits, but afterward you usually cannot start new tasks. The official documentation states that if an account using plan quota or Credits reaches the limit during an active task, that task may continue running to completion; after it finishes, check whether the page offers waiting for recovery, using a reset, adding Credits, or upgrading. Do not assume from this that you can continue indefinitely after exceeding the limit.

Example:A test task is halfway through when the weekly quota reaches zero. It may still return the result for that round; if you then send a new modification request, you may see a quota limit prompt.

First reopen the usage page and confirm the account; a brief display change does not mean a reset has been stably credited. Usage data may take a short time to sync, and client caching or switching accounts or workspaces can also cause apparent inconsistency. Please close and reopen "Settings → Usage," confirm the same account and workspace, and rely on the result after it stabilizes. If it keeps jumping back and forth for a long time, or you still cannot use it after the recovery time, save a screenshot, the client version, the time, and the time zone, and contact OpenAI Support.

Example:On mobile, it shows 100% remaining when first opened, then returns to 8% a few seconds later; do not immediately assume the official reset has happened. Refresh and cross-check on the desktop or at /status.

You can send a follow-up message mid-response to steer Codex immediately. According to the Codex CLI release notes, you can send a follow-up while Codex is responding so it steers immediately, instead of waiting for the current turn to finish.

Example:If Codex is heading toward "write tests first" and you change your mind, just add "skip tests for now, change the interface signature first" and it will pivot right away.

Note:This comes from CLI release notes and covers interaction behavior only—not quota, resets, or plan policy.

No. The homepage predicts public extra resets or card-granting events; your personal account reset time can only be seen on the Usage page. WHENRESET generates candidate times based on public history, recent product phases, and Tibo signals. It answers "when the next public event may occur," not "what time my 5-hour or weekly quota resets," and it cannot read personal accounts.

Example:The homepage countdown still has 2 days left, but your Usage page shows that your 5-hour quota resets tonight at 20:00. Then you only need to wait until tonight at 20:00; you do not need to wait for the homepage countdown to end for your personal reset.

Because predictions are reordered based on new events and new signals; it is not a fixed schedule officially announced. After a new reset or card issuance occurs, historical intervals are recalculated; when Tibo mentions a specific weekday, candidate times temporarily move closer to that signal; when the signal expires without a confirmed event, the page returns to sorting by historical patterns. Consecutive supplementary issuances within the same activity round are also merged to avoid counting them as separate starting points.

Example:On Thursday a post appears saying "there may be a surprise on Friday," so the Friday candidate rises; if Friday ends with no confirmed reset, that signal becomes invalid, and the first candidate may change back to Sunday. This is new information changing the ranking, not an official change to a promise.

No. It only means that among the referenceable historical samples, a certain proportion falls near this candidate time. The number of historical events is limited, and the product stage also changes, so it is not a statistically true probability of occurrence, and even less an official commitment. This number is only used to compare the strength of historical evidence for the current three candidate times.

Example:There are currently 20 historical cycles that still meet the condition of "having waited this long," and 3 of them fall near this candidate time, so it shows about 15%. It only represents "3/20 of the historical distribution" and cannot be interpreted as a real-world 15% probability of occurrence.

Because both give users a new usable opportunity, but the timeline must clearly mark their type and how they take effect. A direct reset refreshes the specified quota for eligible accounts; issuing a card only places a saveable reset into the account, which must be used manually. This site records both, but when calculating predictions it determines whether they belong to the same activity round, to avoid mistakenly treating card issuance, supplementary issuance, and direct reset within a short period as multiple independent cycles.

Example:On September 24 at 10:00, 1 saveable reset is issued first, and at 12:00 it is announced that eligible accounts will be automatically reset. The timeline will keep both entries and mark their types; if they are judged to belong to the same release activity, the prediction calculation will not simply treat them as two independent cycles 2 hours apart.

It mainly syncs Tibo's public posts and distinguishes between "prediction signals" and "events that have already occurred." Mentioning a certain day, a surprise, or capacity improvement can only raise the priority of a candidate time; it is not directly treated as an already completed reset. Only when a post clearly states that a reset has occurred, is being completed and credited, or a saveable reset has been issued, and it is confirmed through system rules, is it written into the official history and a reset notification triggered.

Example:"Tuesday is for surprises" will only be shown as a positive signal for Tuesday; "Reset completed and propagated to eligible users" indicates the reset has been completed, and only then can the event be recorded and configured notifications such as email, mini program, and Feishu group be sent.